在山崩及落石的監測中目前已經有需多成熟的系統，例如影像監控、邊坡防護網而最近也有運用無人機空拍來進行落石災害的巡檢等等。但目前的監測系統中也是受到許多限制，例如夜間監測、受限於植披覆蓋與儀器時間解析度等等限制條件，較難及時地提出正確預警，而地震儀能夠提供補足這些限制的解決方案。然而不同機制例如落石與地震產生的地表震動，會反映在鄰近的地震儀三分量的記錄上，並且其波形特徵與頻率分佈有肉眼能夠區分的明顯差異。
然而在地動訊號分析工作中，山崩、落石所產生的地動訊號搜尋事件的主要方法仰賴主觀的人工判讀挑選，此作法除了耗時，更易造成主觀偏誤，因此本研究試圖利用機器學習的方法將區分落石事件此工作自動化。利用架設於該區的四個地震儀測站之地動訊號事件，我們根據地動訊號大於八倍平均絕對偏差當作其門檻值搜尋到11750筆脈衝訊號當作潛在的未知事件並額外標注了538筆地震事件與524筆落石事件。並以非監督式深度學習模型「深度嵌入聚類演算法」(Deep Embedded Clustering)將訊號依照其訊號本身的相似性進行自動分群，並與手動標籤的地震及落石事件進行比對，驗證模型將地震與落石區分開來的效能。本研究利用地動訊號之時頻圖當作輸入資料集並利用自監督學習「卷積式自編碼模型」(convolutional autoencoder)學習中「編碼器」(encoder) 將時頻圖資料嵌入至64維度顯著之低維特徵中，而DEC演算法有助於學習更有善於聚類效能的時頻圖特徵將訊號依照其類型分成不同類別。而本研究模型的結果顯示能夠將以標籤的落石與地震事件成功區分開，並提升其標籤分類的準確率至97.993%。藉此減少人為標記造成的主觀偏差並大幅度地降低搜尋落石事件的時間，建立出本區域的地動訊號類型特徵，讓我們更加的瞭解本研究區域地動訊號的特性。

In mountain areas, there are often have rockfalls causing traffic problems even safety problems for road users. Therefore, in this study, we use four seismometers on the Luhu, Miaoli, Taiwan from February 25, 2019 to August 31, 2020 to record the 3 component ground motions. Currently, the ground motion recordings contain rockfalls, earthquakes or other natural and anthropogenic sources usually were labeled by people, and this work was a complicated and time-consuming task. In this study, the eight times median absolute deviation (MAD) of the envelope of continuous recordings were used as a threshold for selecting potential seismic events. There are 11,750 qualified events. Among them, 500 and 600 events were manually labeled as earthquakes and rockfalls, respectively. In this study, we use machine learning (ML) techniques for automatically identifying different types of seismic spectrograms. Deep embedded clustering (DEC) is a unsupervised deep clustering algorithm to automatically group different waveforms into classes without manual work. We use spectrograms as input and encode salient features into 64-feature with a convolutional antoencoder. At the same time, the DEC algorithm can learn more friendly clustered 64-feature. In our result, we have five classes to separable types of signals, therefore, earthquake labels were distributed to class0,clss1,class2 and rockfall label were distributed to class4,class5. We evaluated the performance using earthquake and rockfall label. After 1200 iterations, the clustering accuracy improved to 97.993%.

Beaucé, E., Frank, W. B., & Romanenko, A. (2018). Fast matched filter (FMF): An efficient seismic matched‐filter search for both CPU and GPU architectures. Seismological Research Letters, 89(1), 165-172.
Chamarczuk, M., Nishitsuji, Y., Malinowski, M., & Draganov, D. (2020). Unsupervised learning used in automatic detection and classification of ambient‐noise recordings from a large‐N array. Seismological Research Letters, 91(1), 370-389.
Chen, L. (2009). Curse of Dimensionality. In L. Liu & M. T. ÖZsu (Eds.), Encyclopedia of Database Systems (pp. 545-546). Boston, MA: Springer US.
Dammeier, F., Moore, J. R., Haslinger, F., & Loew, S. (2011). Characterization of alpine rockslides using statistical analysis of seismic signals. Journal of Geophysical Research: Earth Surface, 116(F4).
Dietze, M., Mohadjer, S., Turowski, J. M., Ehlers, T. A., & Hovius, N. (2017). Seismic monitoring of small alpine rockfalls – validity, precision and limitations. Earth Surf. Dynam., 5(4), 653-668.
Ester, M., Kriegel, H.-P., Sander, J., & Xu, X. (1996). A density-based algorithm for discovering clusters in large spatial databases with noise. Paper presented at the Proceedings of the Second International Conference on Knowledge Discovery and Data Mining, Portland, Oregon.
Feng, L., Pazzi, V., Intrieri, E., Gracchi, T., & Gigli, G. (2019). Rockfall seismic features analysis based on in situ tests: frequency, amplitude, and duration. Journal of Mountain Science, 16(5), 955-970. doi:10.1007/s11629-018-5286-6
Goodfellow, I., Bengio, Y., Courville, A., & Bengio, Y. (2016). Deep learning (Vol. 1): MIT press Cambridge.
Guo, X., Gao, L., Liu, X., & Yin, J. (2017). Improved Deep Embedded Clustering with Local Structure Preservation. Paper presented at the IJCAI.
Guo, X., Liu, X., Zhu, E., & Yin, J. (2017). Deep clustering with convolutional autoencoders. Paper presented at the International conference on neural information processing.
Hibert, C., Malet, J. P., Bourrier, F., Provost, F., Berger, F., Bornemann, P., . . . Mermin, E. (2017). Single-block rockfall dynamics inferred from seismic signal analysis. Earth Surf. Dynam., 5(2), 283-292.
Hinton, G. E., & Salakhutdinov, R. R. (2006). Reducing the dimensionality of data with neural networks. Science, 313(5786), 504-507.
Holtzman, B. K., Paté, A., Paisley, J., Waldhauser, F., & Repetto, D. (2018). Machine learning reveals cyclic changes in seismic source spectra in Geysers geothermal field. Science advances, 4(5), eaao2929.
Jenkins, W. F., Gerstoft, P., Bianco, M., & Bromirski, P. D. (2021). Unsupervised Deep Clustering of Seismic Data: Monitoring the Ross Ice Shelf. Earth and Space Science Open Archive, 40. doi:10.1002/essoar.10505894.1
Kingma, D. P., & Ba, J. (2014). Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980.
Kong, Q., Trugman, D., Ross, Z., Bianco, M. J., Meade, B., & Gerstoft, P. (2019). Machine Learning in Seismology: Turning Data into Insights. Seismological Research Letters, 90, 3-14.
Krizhevsky, A., Sutskever, I., & Hinton, G. E. (2012). Imagenet classification with deep convolutional neural networks. Advances in neural information processing systems, 25, 1097-1105.
Lacroix, P., Grasso, J. R., Roulle, J., Giraud, G., Goetz, D., Morin, S., & Helmstetter, A. (2012). Monitoring of snow avalanches using a seismic array: Location, speed estimation, and relationships to meteorological variables. Journal of Geophysical Research: Earth Surface, 117(F1).
LeCun, Y., Bottou, L., Bengio, Y., & Haffner, P. (1998). Gradient-based learning applied to document recognition. Proceedings of the IEEE, 86(11), 2278-2324.
Lotti, A., Saccorotti, G., Fiaschi, A., Matassoni, L., Gigli, G., Pazzi, V., & Casagli, N. (2015). Seismic Monitoring of a Rockslide: The Torgiovannetto Quarry (Central Apennines, Italy). In Engineering Geology for Society and Territory - Volume 2 (pp. 1537-1540).
MacQueen, J. (1967). Some methods for classification and analysis of multivariate observations.
Mousavi, S. M., Horton, S. P., Langston, C. A., & Samei, B. (2016). Seismic features and automatic discrimination of deep and shallow induced-microearthquakes using neural network and logistic regression. Geophysical Journal International, 207(1), 29-46.
Mousavi, S. M., Zhu, W., Ellsworth, W., & Beroza, G. (2019). Unsupervised clustering of seismic signals using deep convolutional autoencoders. IEEE Geoscience and Remote Sensing Letters, 16(11), 1693-1697.
Snover, D., Johnson, C. W., Bianco, M. J., & Gerstoft, P. (2020). Deep Clustering to Identify Sources of Urban Seismic Noise in Long Beach, California. Seismological Research Letters, 92(2A), 1011-1022.
Surinach, E., Vilajosana, I., Khazaradze, G., Biescas, B., Furdada, G., & Vilaplana, J. (2005). Seismic detection and characterization of landslides and other mass movements. Natural Hazards Earth Syst. Sci., 5. doi:10.5194/nhess-5-791-2005
Van der Maaten, L., & Hinton, G. (2008). Visualizing data using t-SNE. Journal of machine learning research, 9(11).
Xie, J., Girshick, R., & Farhadi, A. (2016). Unsupervised deep embedding for clustering analysis. Paper presented at the International conference on machine learning.
鍾國平. (2009). 苗栗縣南庄鄉石門及鹿場地滑地穩定性與災害因子之研究. 國立中興大學, Available from Airiti AiritiLibrary database. (2009年)